Anthropic CEO Dario Amodei said in the Financial Times’ “Lunch with the FT” interview that open-source models and Chinese developers could replicate the cybersecurity capabilities of Mythos in 6 to 12 months. Mythos was described as Anthropic’s most capable model and has not been publicly released.
Anthropic has claimed that Mythos previously identified thousands of zero-day vulnerabilities across “every major operating system” and “every major browser.” The company also said the model uncovered a flaw in OpenBSD that had remained hidden for 27 years without being found before. That claim frames Amodei’s warning: he does not believe advanced model capabilities can be locked away forever, and his focus is how many existing vulnerabilities can be patched before replication catches up.
A shorter timeline than earlier internal estimates
Anthropic red-team lead Logan Graham had said earlier that other companies could reproduce Mythos-level capability within months, while Chinese open-source efforts would need 1 to 1.5 years. Amodei’s latest estimate of 6 to 12 months is more aggressive and sits closer to the optimistic end of that range.
His strategic argument is simple. If capability diffusion cannot be stopped, the sequence still matters. Defensive capacity, in his view, should be pushed as far as possible before broad civilian release. Amodei also said he currently treats Mythos like a nuclear weapon: many actors may eventually gain the ability to build it, but access still needs tight control.
Washington and major banks are already assessing risk
The source material says the White House has spent the past two weeks pushing federal agencies to connect to Mythos. Chief of Staff Susie Wiles met directly with Amodei on April 17. At the same time, five major banks — Citi, Goldman Sachs, Bank of America, Morgan Stanley, and JPMorgan Chase — were urgently called in to assess risks to financial infrastructure.
The message from Amodei is less about the race itself and more about the shrinking window before similar capabilities spread wider. His concern is how many holes in existing systems can be closed before others reach the same level.